Differentially Private Conditional Independence Testing
This addresses privacy-preserving statistical analysis for data analysts, offering incremental improvements by adapting known methods to a new privacy setting.
The paper tackles conditional independence testing under differential privacy constraints, designing two private procedures based on existing methods and providing theoretical guarantees and empirical validation, with the first rigorous private tests for continuous Z.
Conditional independence (CI) tests are widely used in statistical data analysis, e.g., they are the building block of many algorithms for causal graph discovery. The goal of a CI test is to accept or reject the null hypothesis that $X \perp \!\!\! \perp Y \mid Z$, where $X \in \mathbb{R}, Y \in \mathbb{R}, Z \in \mathbb{R}^d$. In this work, we investigate conditional independence testing under the constraint of differential privacy. We design two private CI testing procedures: one based on the generalized covariance measure of Shah and Peters (2020) and another based on the conditional randomization test of Candès et al. (2016) (under the model-X assumption). We provide theoretical guarantees on the performance of our tests and validate them empirically. These are the first private CI tests with rigorous theoretical guarantees that work for the general case when $Z$ is continuous.